[news.eclipse.technology.ohf] Re: stochastic simulations & multi-cpu env

["Daniel Ford" <daford@xxxxxxxxxxxxxxx>]

Vincent,
    here's the long story.....the Eclipse version of STEM is the 2nd 
implementation of STEM.  The first version was released on alphaWorks about 
two years ago.  That version had both deterministic and stochastic 
implementations for SI, SIR and SEIR models.  The implementations in the new 
Eclipse code base are the same mathematics as in the first.

I've been very careful with testing the implementations of the disease model 
mathematics.  I created simple scenarios and worked out all the computations 
by hand to make sure that they were correct (otherwise, what is the point?) 
I only had enough cycles to complete those calculations for deterministic 
calculations, not for stochastic.  So, because they were not tested, they 
were not exposed in this first milestone release.

We have a number of different kinds of disease models in plan that will be 
considerably more sophisticated than what currently exists in the code base. 
Some are models we're working on internally and some are ones we've funded 
others to develop (e.g., 
http://web.mit.edu/newsoffice/2007/influenza-0531.html).  We should probably 
start up a thread on the topic.

As for multiprocessor environments, I think Eclipse is at the mercy of the 
JVM and its ability to influence threading (not much).  STEM uses the 
Eclipse Jobs api for threading and is completely at the mercy of Eclipse's 
scheduling of threads to Jobs.   In the current implementation, there is a 
single thread to compute the next state of the canonical graph.  I'm 
skeptical that this could change easily as there is a specified order for 
execution of the Decorators defined in the tree that defines the Model of 
the Scenario.  The graphics for the Map view delegate the rendering of the 
image to a background job and all of the I/O to retrieve lat/long data is 
parallelized by multiple Jobs, one for each country.

Dan

"Vincent David" <vega42@xxxxxxxxx> wrote in message 
news:f53tca$81d$1@xxxxxxxxxxxxxxxxxxxx
> Hi,
>
> since I got this thing running meanwhile I do have some more specific 
> questions I would like to address here. I did have a look in the docs and 
> code but I do undestand that you guys are still quite deep in the 
> development phase, so I will ask you directly.
>
> 1. How far are the stochastic models actually implemented? I find a basic 
> framework in stem.diseaesemodel but have the impresion, that those haven't 
> been implemented yet. Is this correct?
>
> 2. What about multiprocesser-environments? Is there an easy way to 
> instruct Eclipse to thread over multiple cores of one engine? What about 
> clusters?
>
> thx
>
> Vincent